Physics 212: Statistical mechanics II, Fall 2005 Lecture XX

In the last lecture we used the Flory model of the self-avoiding walk (SAW) to obtain an estimate of the exponent that governs the typical size of a polymer: R N, = 3 . d+2 (1)

However, we still have little understanding of the polymer exponents and , or why these quantities might be universal. The main goal of the rst part of this lecture is to develop a connection between self-avoiding walks and critical points of spin models. Then all the approaches and insights previously developed for spin models can be directly applied to the polymer problems. Note that there is no energy or Hamiltonian in the original self-avoiding walk problem. Recall the good old Ising model on a lattice, whose partition function is (again the T r means a sum over all spin congurations) ZI = T r e Now this can be rewritten using eKsi sj = cosh K + si sj sinh K, which holds since either si sj = 1 or si sj = 1 on each bond. So we have ZI = (cosh K)N T r
ij
ij

Ksi sj

= Tr
ij

eKsi sj .

(2)

(3)

(1 + si sj tanh K),

(4)

where N is the number of sites. The prefactor is a smooth function of K for any N , so it cannot aect the critical properties, and will be dropped (it can be restored at the end if desired). The rest takes the simple form in terms of v = tanh K ZI = T r
ij

(1 + vsi sj ),

(5)

Now that ZI is in this convenient form, let us carry out an expansion for small K (high temperature). All we have to do is write out the term of order v N , then trace over the si variables to get the result. However, we see immediately that most of these terms vanish. For instance, the linear in v term is Tr vsi sj = 0. (6)
ij

Any term which has one of the spin variables appearing an odd number of times will be equal to zero, since the trace over that spins values 1 will cancel. Hence the nonvanishing terms are those in which every spin variable occurs an even number of times. To make things easier, let us 1

put our Ising model on the hexagonal lattice, in which each spin has three nearest neighbors. The point of doing this is that each spin variable can appear either zero times, one time, two times, or three times. So the only possible nonzero terms have, for each spin si , either s2 or 1. (For the i square lattice, we would also have to include s4 , as there are four bonds per site: this means that i intersections of loops are allowed, and the connection to polymers is more complicated.) The terms in the partition function ZI in (5) can be represented graphically as follows. For every bond that appears with coecient v, we draw a solid line on the lattice. Then terms which do not trace to zero must correspond to sets of closed nonselntersecting paths on the lattice. The paths must be nonintersecting since any self-intersection on the hexagonal lattice requires three bonds to go into one site, so the overall term will trace to zero. Thus we can write schematically ZI =
m

v m (sum of closed loops with total length m).

(7)

Here the Ising model is especially simple since, for a nonzero term like s2 s2 s2 s2 involving m spins, 1 2 3 4 its value is just 1 independent of the spins. Hence ZI = 2N
m

v m (number of sets of loops with m total bonds).

(8)

This factor 2N has no K dependence and can be ignored if only critical phenomena are desired. Restoring the dropped prefactors, we obtain the so-called loop gas representation: ZI = (2 cosh K)N
G

vm.

(9)

In this standard notation, G is a group of nonintersecting loops of total length m. Lets see how this generalizes if, instead of the Ising model, we consider the O(n) model, i.e., the local spin on the lattice is a unit vector in n dimensions. One slight dierence from what weve done before is that, instead of introducing a Hamiltonian, we just redene the expression for the partition function (5): Zn = T r (1 + vsi sj ). (10) Here weve ignored possible smooth prefactors, and it is not obvious that this Zn is what would be obtained starting from the exponential of the O(n) Hamiltonian we used before. The approach in polymer physics is to dene the O(n) model through (10), and guess that this turns out to lie in the same universality class as what we studied before. Each spin is integrated over the surface of an n-dimensional sphere of radius n (this normalization is chosen to get each components average length to be 1) . It is convenient to pull out a factor of the volume of the sphere: the trace over one spin, written out as an integral over the sphere, is T r si = di = Sn di (11)

N where now the normalized volume element d = d/Sn integrates to 1. The overall prefactor Sn N in the Ising case) will be ignored. (analogous to the 2

The main dierence in the O(n) case, which is very important, will be that dierent congurations of total length m will have dierent weights. Again, by symmetry, the only nonzero terms will occur when each spin occurs an even number of times. Note that, by the symmetry of the sphere, si si di = . (12)

(There is no 1/n on the right side because of the normalization for i s2 = n chosen above.) Here i si1 , si2 , . . . are the n components of spin si . This implies that not just each spin i must appear an even number of times (zero or two), but each component six must appear an even number of times. So, for instance, terms like six siy will trace to zero, even though spin i appears twice. Hence, since the interaction is diagonal in components, all the same components must appear in a closed loop. However, when there are multiple loops, dierent components can appear in dierent loops. That means that we have to add a weighting factor nL , where n is the number of components and L is the number of dierent loops. Thus the loop gas representation of the O(n) model is the famous formula (ignoring smooth prefactors) Zn =
G

v m nL .

(13)

Here G is a group of disjoint loops on the honeycomb lattice, m is the total number of bonds, and L is the number of loops in G. This formula is useful because, in this loop representation, we can analytically continue the model to nonintegral values of n. All n controls is the weighting of dierent terms in the loop sum, and these weights change continuously as the real number n changes continuously. In particular, we can take n 0 so that the sum over groups of loops G becomes a sum over single loops. More precisely, dene the quantity (this basically the free energy per site, up to a T factor we ignore) f (n) = lim 1 log Zn , N N (14)

so that f (n) is nite even as N gets large. You can verify that g(n) lim f (n) vm, = n SAP (15)

n0

where the sum is over all self-avoiding polygons of length m, i.e., closed nonselntersecting loops on the lattice, with the degeneracy from translations canceled out. That is, each polygon only appears once in the above sum, while before in Zn the same polygon would appear, shifted to all dierent points of the lattice. Now were ready to use the above expression. The logarithm of the partition function of the O(n) model per site is a Taylor series in v, where the coecient of each term v m is the number of SAPs of length m, which in our previous notation is qm . We believe that qm should go as qm m m3 , (16)

and want to connect the quantities and to the O(n) Taylor series for the free energy (ignoring prefactors) g(n) = qm v m = (v)m m3 . (17) 3

Immediately we see that this sum has a special point at v = 1. To investigate what happens near there, note that if = 3 we would just get the Taylor series for 1 . 1 v You can check that in general the leading singularity of the above as v 1 is g(n) (1 v)2 = t2 , where here t |1 v| is again the distance from the special point v = 1. Finally, the reason why we dened in this way can now be explained. The free energy in the spin model, which has the same singularity as g, should be singular as t2 , since two derivatives with respect to t of the free energy give the specic heat, which should go as t . So that appeared in qm is the same as the specic heat exponent of the O(n) model as n 0. This connection was originally derived by de Gennes, whose book on polymer physics is highly recommended (it includes also the less mathematical areas of the eld). High-temperature expansions like those discussed above have other uses besides polymers, as mentioned earlier. These high-temperature expansions have been carried out to amazingly high order using computers, and can be used to provide accurate numerical estimates of critical exponents for comparison to -expansion predictions and other methods. Before proceeding to understand where the other polymer exponents come from, we comment briey on what realistic interactions correspond to the self-avoiding walk model. Suppose that, in addition to the excluded-volume repulsive interaction, an attractive interaction between the polymer elements is turned on. For weak attraction, the polymer behaves like an SAW (for example, in 2D R N 3/5 ). For strong attraction, the polymer goes into a clumped or dense phase with R N 1/d . The clumped and dense phases are separated by one special point, called the -point, which has R N 4/7 in 2D. The -point that separates the extended SAW phase from the clumped phase corresponds to a critical point of the attractive interaction, or a tricritical point in the spin language, since ordinary polymer problems with no tuning correspond to critical points in the spin language. Actually the tuning parameter in the SAW problem should be thought of as the innite length of the polymer: the nite length of a long polymer represents a small perturbation around the xed point at innite length. Percolation is another geometrical critical phenomenon similar to SAWs, which we may not have time to discuss in this course. It can also be related exactly to spin-like models: percolation is the q 1 limit of the q-state Potts model, which is a lattice model with q states per site and an interaction between sites i and j proportional to qi ,qj . (19) (18)